Most validation work on DNA methylation (DNAm) clocks has focused on their ability to predict future mortality or disease, not on whether they actually move in response to an intervention. A recent large-scale analysis addressed this directly by harmonizing 51 longevity interventional studies with pre- and post-intervention DNA methylation data, calculating 16 core epigenetic clocks plus dozens of additional biomarkers across all of them. The headline result: 19 of the 51 interventions significantly decreased epigenetic age across the biomarker panel, 5 significantly increased it, and 26 showed no significant effect — so responsiveness is real, but far from universal.
The more important finding is which clocks actually track change well. Newer, reliability-optimized "Generation 2+" clocks — GrimAge2, PC-GrimAge, PC-PhenoAge, SystemsAge, and especially DunedinPACE — showed consistent, significant decreases across interventions, whereas the original first-generation clocks (Horvath1, Horvath2, Hannum) showed scattered, inconsistent changes with no clear pattern. DunedinPACE stood out as the single most responsive biomarker, decreasing significantly in 16 of the interventions studied and increasing in only one, while also showing the best agreement with other clocks when it detected an effect. This matters practically: pace-of-aging and reliability-optimized clocks (built using only CpGs with strong test-retest reliability, or using principal-component versions of older clocks) appear to be far better suited as trial endpoints than the original chronological-age predictors.
Intervention type matters a lot
Pharmacological interventions produced the largest effects. Across the dataset, pharmacological interventions (metformin, anti-TNF therapies, semaglutide, rapamycin, and others) drove significantly larger decreases in epigenetic age than lifestyle, supplement, or medical-procedure interventions, with anti-TNF therapy and metformin showing especially consistent effects. Lifestyle interventions (diet, exercise) also produced significant average decreases, just smaller in magnitude. Supplements and medical procedures as categories did not reach significance overall, though individual studies within those categories sometimes did.
Consistency across replications is itself informative. Anti-TNF therapy and Mediterranean-diet studies modified the same set of biomarkers in the same direction across independent studies, supporting them as bona fide interventions, whereas senolytic trials (dasatinib+quercetin) and certain supplement studies showed biomarkers moving in inconsistent or even opposite directions across replications — a caution against over-interpreting single-study senolytic results.
Health status of the study population is a major moderator
One of the more consequential findings: DNAm biomarkers were far more responsive in disease populations than in healthy ones — multiple Gen 2+ biomarkers showed significant decreases in disease cohorts, while only a handful (notably DunedinPACE and PC-GrimAge) showed any significant response in healthy populations. Practically, this means trials in healthy, already-aging-well volunteers (the population most longevity-supplement studies enroll) are working against a much smaller effect size and need larger samples or more sensitive endpoints (DunedinPACE again performing best here) to detect anything.
What individual landmark trials show
A few well-known RCTs illustrate the same inconsistency-across-clocks pattern:
- CALERIE (calorie restriction): of 11 biomarkers tested, only DunedinPACE and PhenoAge decreased significantly, and the PhenoAge result was undercut by its more reliable counterpart (PC-PhenoAge) not reaching significance — again pointing to DunedinPACE as the most trustworthy signal.
- DO-HEALTH (vitamin D + omega-3 + exercise in older adults): this 2025 Nature Aging trial found individual and additive effects of the three interventions on DNAm clocks, one of the better-powered multi-arm RCTs in this space.
- DAMA (diet + exercise): GrimAge decreased by 0.66 years; only GrimAge was measured, limiting cross-comparison.
- Smaller supplement and vitamin D studies mostly showed effects in only one clock out of several tested, or only in subgroups — a pattern consistent with either true clock-specific sensitivity or publication-bias-prone single-clock reporting.
- A 2026 RCT of a yogurt-probiotic-plus-lifestyle intervention specifically tested DunedinPACE as a short-term-responsive marker, reflecting a growing consensus that pace-of-aging clocks can register change over weeks to months, unlike first-generation clocks which usually need longer follow-up.
Mechanistic/system-specific clocks add resolution
Newer "explainable" clocks that break aging into organ-system or protein/metabolite components (SystemsAge, OMICmAge) can detect effects that whole-body clocks miss. For example, gastric bypass surgery and a lavender-oil supplement showed no significant change in any general Gen2+ clock, but did show significant decreases in system-specific scores — gastric bypass in the metabolic score, lavender oil across brain, immune, inflammation, kidney, and metabolic scores — while smoking cessation's effect concentrated specifically in the lung-aging score and metformin's effect spread across inflammation, brain, and metabolic scores. This suggests whole-body composite clocks can under-detect interventions with narrow, organ-specific mechanisms.
.png)
Bottom line
- Yes, epigenetic biomarkers are responsive to longevity interventions — but responsiveness is clock-dependent, not universal across the whole DNAm-clock ecosystem.
- Reliability-optimized, "pace of aging" clocks (especially DunedinPACE) and PC-transformed Gen2+ clocks are the most trustworthy and sensitive endpoints; original Horvath/Hannum-style clocks are noisy and inconsistent for this purpose.
- Pharmacological interventions (anti-TNF, metformin, GLP-1s) show the strongest and most reproducible signals, followed by lifestyle interventions; supplement and single-procedure studies are weaker and less consistent.
- Disease populations show much larger, more detectable responses than healthy populations — an important consideration for trial design and sample-size planning.
- Regulatory acceptance is still the bottleneck: even with growing responsiveness data, epigenetic clocks are not yet accepted as validated surrogate endpoints for drug approval, largely because it hasn't yet been established that a clock's movement actually predicts a change in real clinical outcomes (mortality, disease incidence) — that causal link is the next major evidence gap the field needs to close.